Accurate endoscopic assessment of the depth of submucosal invasion in T1 colorectal cancer is critical for determining the optimal treatment strategy. This assessment is currently performed with narrow-band imaging (NBI) and magnifying chromoendoscopy (MCE), yet the superiority of either modality in direct comparisons remains unclear. We therefore restricted our evaluation to paired, within-study direct comparative diagnostic accuracy studies in which both NBI and MCE were applied to the same lesion and compared the diagnostic performance of NBI and MCE for identifying deep submucosal invasion (T1b).

We systematically searched PubMed, Embase, and Cochrane CENTRAL from inception to September 19, 2025 using MeSH/keywords for NBI, narrow-band imaging, JNET, magnifying chromoendoscopy, Kudo classification, pit pattern, deep submucosal invasion, T1 colorectal cancer, T1b. We included paired, within-study direct comparative diagnostic accuracy studies in adults in which both NBI and MCE were applied to the same lesion during the same session, 2×2 data for each test could be reconstructed and histology was the reference standard with T1b defined as ≥1,000 µm (SM2/3). Two reviewers independently screened and extracted data; risk of bias was assessed with QUADAS-2 and QUADAS-C. The primary synthesis used a Bayesian bivariate HSROC model with test type as a covariate to estimate pooled sensitivity/specificity and relative effects (ratios and differences; NBI/MCE), reported as posterior medians with 95% credible intervals. Small-study effects were examined with Deeks’ test (all p > 0.10).

Eleven studies (7 prospective, 4 retrospective; 2928 lesions) were included. Pooled performance: NBI sensitivity 0.746 (95% CrI 0.629 to 0.834) and specificity 0.962 (0.930 to 0.981); MCE sensitivity 0.817 (0.728 to 0.890) and specificity 0.941 (0.892 to 0.967). Comparative effects (NBI vs MCE) showed no clear differences: sensitivity ratio 0.914 (95% CrI 0.744 to 1.069; difference −0.071, −0.218 to 0.052) and specificity ratio 1.021 (0.982 to 1.077; difference 0.020, −0.017 to 0.069). Heterogeneity was moderate–substantial; index-test risk of bias was common on QUADAS-C; small-study effects were not evident (Deeks p = 0.30).

In paired within-study comparative analyses, NBI and MCE showed comparable diagnostic accuracy for identifying T1b, with substantial overlap of their 95% credible intervals. At present, an NBI-first strategy with selective MCE add-on is reasonable for the initial assessment of colorectal lesions.